Segmentation-guided multi-modal brain tumors survival prediction model using pseudo-labeling approach.

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Machine learning in healthcare

Background:

• Accurate brain tumor survival prediction is crucial for treatment planning and patient prognosis.
• Current deep learning methods often require multiple networks and neglect population data, while censored data poses challenges.
• Existing models struggle with suboptimal performance due to incomplete patient survival information.

Purpose of the Study:

• To develop an advanced deep learning framework for precise brain tumor survival prediction.
• To enhance the utilization of population information and address challenges associated with censored survival data.
• To create a novel dataset for brain tumor segmentation and survival prediction.

Main Methods:

• Proposed an end-to-end multi-model pseudo-label approach for survival prediction.
• Integrated patient population information to optimize predictive model performance.
• Developed a novel class label generation method to enlarge sample size and improve data utilization.
• Utilized and supplemented the BraTS 2021 dataset for segmentation and survival prediction tasks.

Main Results:

• The proposed model demonstrated enhanced precision in predicting brain tumor patient survival rates.
• Experimental results confirmed the model's superior generalization capabilities compared to existing methods.
• The integrated approach effectively handled censored data, improving prediction accuracy.

Conclusions:

• The developed multi-model pseudo-label approach offers a significant advancement in brain tumor survival prediction.
• Integrating population data and novel data augmentation techniques improves model performance and generalization.
• The new dataset and methodology pave the way for more accurate clinical decision-making in neuro-oncology.